On the Adequacy of GIS-Generated Weed Maps for Precision Farming

Discrete sampling data is used in several environmental studies to create maps in order to support decision-making processes. The decision maps represent an increasing importance in modern Precision Farming. For the application of herbicides in a field, maps of weed distribution are necessary. The uncertainty of those, resulting from sparse sampling patterns is one major reason why farmers tend to be hesitant in applying the GIS-generated weed maps. In this paper, the lack of predictability and the problems of weed maps will be exemplified. For that purpose, approximately 2800 pictures of the surface of the ground that were taken on a maize field could be manually evaluated and compared to a well-established sampling grid for weeds using cross-correlation. The results prove low values of correlation between the produced weed maps and emphasise the uncertainty of these maps by a simulation experiment which will be described later on in this paper.